3 ms·
It seems that a lot of the leaders are the results of mixing finetunes, which really makes me think that there was a leak of test sets into the training data.
by SubiculumCode 3y ago
It seems that a lot of the leaders are the results of mixing finetunes, which really makes me think that there was a leak of test sets into the training data.
- SubiculumCode 3y agoTo reply to myself. I am not saying that this model did, or that even if it did, that it was done intentionally. ML is hard, and there are so many ways for data to leak. What I AM surprised about is that it is not clear what CultriX did that was better than what a ton of others have done. Any clues?